Update README.md
Browse files
README.md
CHANGED
|
@@ -3,14 +3,32 @@ license: apache-2.0
|
|
| 3 |
base_model:
|
| 4 |
- NovaSky-AI/Sky-T1-32B-Preview
|
| 5 |
library_name: peft
|
|
|
|
|
|
|
| 6 |
---
|
| 7 |
|
| 8 |
# Sky-T1-32B with Step-by-Step Thinking
|
| 9 |
|
| 10 |
-
This repo contains a LoRA adapter to [NovaSky-AI/Sky-T1-32B-Preview](https://huggingface.co/NovaSky-AI/Sky-T1-32B-Preview) for step-by-step thinking.
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
Each step is outputted in a specific format:
|
| 13 |
```
|
| 14 |
<|step|> TITLE
|
| 15 |
TEXT
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
```
|
|
|
|
| 3 |
base_model:
|
| 4 |
- NovaSky-AI/Sky-T1-32B-Preview
|
| 5 |
library_name: peft
|
| 6 |
+
datasets:
|
| 7 |
+
- shakedzy/Sky-T1_data_steps
|
| 8 |
---
|
| 9 |
|
| 10 |
# Sky-T1-32B with Step-by-Step Thinking
|
| 11 |
|
| 12 |
+
This repo contains a LoRA adapter to [NovaSky-AI/Sky-T1-32B-Preview](https://huggingface.co/NovaSky-AI/Sky-T1-32B-Preview) for step-by-step thinking.
|
| 13 |
+
It breaks down the thinking process to steps, allowing better tracking of the thought process.
|
| 14 |
+
|
| 15 |
+
The model was finetuned without any system prompt.
|
| 16 |
|
| 17 |
Each step is outputted in a specific format:
|
| 18 |
```
|
| 19 |
<|step|> TITLE
|
| 20 |
TEXT
|
| 21 |
+
```
|
| 22 |
+
Thus the model's response should adhere to the following format:
|
| 23 |
+
```
|
| 24 |
+
<|begin_of_thought|>
|
| 25 |
+
<|step|> STEP_1_TITLE
|
| 26 |
+
STEP_1_THINKING
|
| 27 |
+
<|step|> STEP_2_TITLE
|
| 28 |
+
STEP_2_THINKING
|
| 29 |
+
...
|
| 30 |
+
<|end_of_thought|>
|
| 31 |
+
<|begin_of_solution|>
|
| 32 |
+
...
|
| 33 |
+
<|end_of_solution|>
|
| 34 |
```
|